{"id":"W3191474316","doi":"10.3390/dj9080094","title":"Evaluating Classification Consistency of Oral Lesion Images for Use in an Image Classification Teaching Tool","year":2021,"lang":"en","type":"article","venue":"Dentistry Journal","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Faculty of Medicine and Dentistry, University of Alberta; University of Alberta","keywords":"Kappa; Rubric; Consistency (knowledge bases); Artificial intelligence; Cohen's kappa; Medicine; Inter-rater reliability; Lesion; Pattern recognition (psychology); Computer science; Mathematics; Machine learning; Statistics; Pathology; Mathematics education","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02745558,0.0005961097,0.0006060555,0.004997215,0.0006828613,0.001378284,0.001175536,0.0009581503,0.001419193],"category_scores_gemma":[0.1060641,0.0003690381,0.0007388981,0.001379627,0.0008775737,0.001253758,0.001861041,0.0006156758,0.0006468968],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007787358,"about_ca_system_score_gemma":0.000657953,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001300168,"about_ca_topic_score_gemma":0.002487361,"domain_scores_codex":[0.9846871,0.005069455,0.002781538,0.002159737,0.004816868,0.0004853019],"domain_scores_gemma":[0.8500164,0.1014551,0.01059275,0.008413272,0.02792151,0.001600935],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00382929,0.001442554,0.5206711,0.001254297,0.0005587937,0.0009077005,0.02043684,0.004219044,0.06484244,0.0007230735,0.004010133,0.3771047],"study_design_scores_gemma":[0.0001711313,0.003413107,0.8709365,0.0004043198,0.000359154,0.002596083,0.009143295,0.04714053,0.05829221,0.0009642826,0.006314249,0.0002651798],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9717745,0.0002743731,0.02385629,0.0001022515,0.0000768656,0.0007110144,0.0002422674,0.0005016451,0.002460801],"genre_scores_gemma":[0.9582254,0.0001596566,0.03987649,0.00004792705,0.00002688579,0.0003807631,0.0003850893,0.0001333637,0.0007643882],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02745558,"threshold_uncertainty_score":0.1452007,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2608180206976614,"score_gpt":0.490187083475301,"score_spread":0.2293690627776396,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}